Hello! 👋
It’s Thursday, 17th September 2026. Hello and welcome back to Bold Efforts!
Let us go back 161 years. In 1865, a British economist named William Stanley Jevons noticed something that did not make much sense.
Steam engines were getting better.
Newer engines could produce the same amount of power while burning less coal. The obvious conclusion was that Britain would eventually need less coal.
The opposite happened. Coal consumption exploded.
The better steam engines became, the more useful they became. Steam power got cheaper, so factories used more of it. New industries adopted it. More engines were built. Each engine needed less coal to do the same job, but there were so many more engines doing so many more things that Britain ended up consuming more coal, not less.
It became known as Jevons Paradox.
I have been thinking about this because we may be running the same experiment again. Except this time, the resource getting cheaper is intelligence.
Most conversations about AI and jobs contain a hidden assumption. There is a fixed amount of work to be done. In fact, recently while meeting friends, most of them agreed with this. I have an unpopular opinion.
A company has 100 units of work. Ten people currently do it. AI makes each person twice as productive. Therefore, the company now needs five people. Fair enough so far.
But what happens if AI makes something so much cheaper that we suddenly want far more of it?
I notice this in my own work already. If answering a question properly would take me half a day, I have to decide whether the question deserves half a day.
Most do not. But if AI reduces that cost to twenty minutes, I ask the question. Then I ask another.
Then I test a different assumption. Then I look at three competitors instead of one. Then I run the analysis again from another angle because now that costs almost nothing too. So I ask a model the same question from 10 different angles to arrive at a slightly better answer every time.
I did not use AI to do the same amount of research in less time. I used it to justify doing research that I would never have done before. I guess most of us work similarly.
Imagine building a small piece of internal software used to require three engineers and six weeks.
A company probably had a long list of things that would have been useful to build, but were simply not useful enough to justify the cost.
Now suppose AI cuts that cost by 80%. The company could use the productivity gain to employ fewer engineers.
Or it could finally build all the things that never made it past the spreadsheet. Both are perfectly rational outcomes.
The same thing can happen almost everywhere. When analyzing every customer conversation is expensive, you analyze a sample.
When it becomes cheap, you analyze all of them. The map of a territory becomes map of the world so to say.
When creating ten versions of a campaign is expensive, you create two. When it becomes cheap, you create fifty and test them.
When researching a tiny market is expensive, you ignore it. When an AI agent can do the first pass in minutes, suddenly the question is worth asking.
This is the part of the AI and jobs debate that I think we underweight. There is a huge amount of latent work in the world.
Questions nobody researches. Software nobody builds. Customers nobody serves individually. Ideas nobody tests. Processes nobody audits closely.
Because until now, their value was lower than their cost. AI changes that equation. And once the equation changes, demand can change with it.
This does not mean Jevons Paradox will magically save every job.
Some demand really is fixed. A company does not need ten times as many payroll runs because payroll becomes easier. You do not need twenty annual tax filings because an accountant becomes twenty times more productive.
In those areas, automation can simply mean fewer hours of human labour.
And even where total demand grows, there is no guarantee that it benefits the people whose tasks were automated. New work can appear somewhere completely different.
Jevons Paradox is not a worker-level insurance policy. It is a reminder that economies are dynamic.
When the price of something falls dramatically, people change their behaviour. That is what makes the next few years interesting.
We may end up in a world where AI does an extraordinary amount of work and humans somehow have more work around them than before.
More experiments become worth running. More products become worth building. More decisions become worth analyzing. More niches become worth serving.
More ideas become cheap enough to try. And expectations rise with the new capability.
Once software becomes cheap enough to build for tiny workflows, people start wondering why those workflows still happen manually.
Efficiency does not always buy leisure. Sometimes it simply moves the baseline.
That might be the real paradox of AI and work. We keep imagining a future where machines do more, so there must be less left for us to do.
History offers another possibility. Machines do more. The cost of doing things collapses. And suddenly we discover an enormous number of things that were never worth doing before.
The future of work may not contain less work at all.
It may contain a much larger definition of what is worth doing.
Thank you for reading. See you next week.
Best,
Kartik
I write Bold Efforts every week to think clearly about where work and life are actually headed. If you want these essays in your inbox, you can subscribe here.